1 /* $NetBSD: kern_event.c,v 1.55 2008/04/22 12:04:22 ad Exp $ */ 2 3 /*- 4 * Copyright (c) 2008 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. All advertising materials mentioning features or use of this software 16 * must display the following acknowledgement: 17 * This product includes software developed by the NetBSD 18 * Foundation, Inc. and its contributors. 19 * 4. Neither the name of The NetBSD Foundation nor the names of its 20 * contributors may be used to endorse or promote products derived 21 * from this software without specific prior written permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 24 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 25 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 26 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 27 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 28 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 29 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 30 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 31 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 32 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 33 * POSSIBILITY OF SUCH DAMAGE. 34 */ 35 36 /*- 37 * Copyright (c) 1999,2000,2001 Jonathan Lemon <jlemon@FreeBSD.org> 38 * All rights reserved. 39 * 40 * Redistribution and use in source and binary forms, with or without 41 * modification, are permitted provided that the following conditions 42 * are met: 43 * 1. Redistributions of source code must retain the above copyright 44 * notice, this list of conditions and the following disclaimer. 45 * 2. Redistributions in binary form must reproduce the above copyright 46 * notice, this list of conditions and the following disclaimer in the 47 * documentation and/or other materials provided with the distribution. 48 * 49 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 50 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 51 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 52 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 53 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 54 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 55 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 56 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 57 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 58 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 59 * SUCH DAMAGE. 60 * 61 * FreeBSD: src/sys/kern/kern_event.c,v 1.27 2001/07/05 17:10:44 rwatson Exp 62 */ 63 64 #include <sys/cdefs.h> 65 __KERNEL_RCSID(0, "$NetBSD: kern_event.c,v 1.55 2008/04/22 12:04:22 ad Exp $"); 66 67 #include <sys/param.h> 68 #include <sys/systm.h> 69 #include <sys/kernel.h> 70 #include <sys/proc.h> 71 #include <sys/file.h> 72 #include <sys/select.h> 73 #include <sys/queue.h> 74 #include <sys/event.h> 75 #include <sys/eventvar.h> 76 #include <sys/poll.h> 77 #include <sys/malloc.h> /* for hashinit */ 78 #include <sys/kmem.h> 79 #include <sys/stat.h> 80 #include <sys/filedesc.h> 81 #include <sys/syscallargs.h> 82 #include <sys/kauth.h> 83 #include <sys/conf.h> 84 #include <sys/atomic.h> 85 86 static int kqueue_scan(file_t *, size_t, struct kevent *, 87 const struct timespec *, register_t *, 88 const struct kevent_ops *, struct kevent *, 89 size_t); 90 static int kqueue_ioctl(file_t *, u_long, void *); 91 static int kqueue_fcntl(file_t *, u_int, void *); 92 static int kqueue_poll(file_t *, int); 93 static int kqueue_kqfilter(file_t *, struct knote *); 94 static int kqueue_stat(file_t *, struct stat *); 95 static int kqueue_close(file_t *); 96 static int kqueue_register(struct kqueue *, struct kevent *); 97 static void kqueue_doclose(struct kqueue *, struct klist *, int); 98 99 static void knote_detach(struct knote *, filedesc_t *fdp, bool); 100 static void knote_enqueue(struct knote *); 101 static void knote_activate(struct knote *); 102 103 static void filt_kqdetach(struct knote *); 104 static int filt_kqueue(struct knote *, long hint); 105 static int filt_procattach(struct knote *); 106 static void filt_procdetach(struct knote *); 107 static int filt_proc(struct knote *, long hint); 108 static int filt_fileattach(struct knote *); 109 static void filt_timerexpire(void *x); 110 static int filt_timerattach(struct knote *); 111 static void filt_timerdetach(struct knote *); 112 static int filt_timer(struct knote *, long hint); 113 114 static const struct fileops kqueueops = { 115 (void *)enxio, (void *)enxio, kqueue_ioctl, kqueue_fcntl, kqueue_poll, 116 kqueue_stat, kqueue_close, kqueue_kqfilter 117 }; 118 119 static const struct filterops kqread_filtops = 120 { 1, NULL, filt_kqdetach, filt_kqueue }; 121 static const struct filterops proc_filtops = 122 { 0, filt_procattach, filt_procdetach, filt_proc }; 123 static const struct filterops file_filtops = 124 { 1, filt_fileattach, NULL, NULL }; 125 static const struct filterops timer_filtops = 126 { 0, filt_timerattach, filt_timerdetach, filt_timer }; 127 128 static u_int kq_ncallouts = 0; 129 static int kq_calloutmax = (4 * 1024); 130 131 MALLOC_DEFINE(M_KEVENT, "kevent", "kevents/knotes"); /* for hashinit */ 132 133 #define KN_HASHSIZE 64 /* XXX should be tunable */ 134 #define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask)) 135 136 extern const struct filterops sig_filtops; 137 138 /* 139 * Table for for all system-defined filters. 140 * These should be listed in the numeric order of the EVFILT_* defines. 141 * If filtops is NULL, the filter isn't implemented in NetBSD. 142 * End of list is when name is NULL. 143 * 144 * Note that 'refcnt' is meaningless for built-in filters. 145 */ 146 struct kfilter { 147 const char *name; /* name of filter */ 148 uint32_t filter; /* id of filter */ 149 unsigned refcnt; /* reference count */ 150 const struct filterops *filtops;/* operations for filter */ 151 size_t namelen; /* length of name string */ 152 }; 153 154 /* System defined filters */ 155 static struct kfilter sys_kfilters[] = { 156 { "EVFILT_READ", EVFILT_READ, 0, &file_filtops, 0 }, 157 { "EVFILT_WRITE", EVFILT_WRITE, 0, &file_filtops, 0, }, 158 { "EVFILT_AIO", EVFILT_AIO, 0, NULL, 0 }, 159 { "EVFILT_VNODE", EVFILT_VNODE, 0, &file_filtops, 0 }, 160 { "EVFILT_PROC", EVFILT_PROC, 0, &proc_filtops, 0 }, 161 { "EVFILT_SIGNAL", EVFILT_SIGNAL, 0, &sig_filtops, 0 }, 162 { "EVFILT_TIMER", EVFILT_TIMER, 0, &timer_filtops, 0 }, 163 { NULL, 0, 0, NULL, 0 }, 164 }; 165 166 /* User defined kfilters */ 167 static struct kfilter *user_kfilters; /* array */ 168 static int user_kfilterc; /* current offset */ 169 static int user_kfiltermaxc; /* max size so far */ 170 static size_t user_kfiltersz; /* size of allocated memory */ 171 172 /* Locks */ 173 static krwlock_t kqueue_filter_lock; /* lock on filter lists */ 174 static kmutex_t kqueue_misc_lock; /* miscellaneous */ 175 176 /* 177 * Initialize the kqueue subsystem. 178 */ 179 void 180 kqueue_init(void) 181 { 182 183 rw_init(&kqueue_filter_lock); 184 mutex_init(&kqueue_misc_lock, MUTEX_DEFAULT, IPL_NONE); 185 } 186 187 /* 188 * Find kfilter entry by name, or NULL if not found. 189 */ 190 static struct kfilter * 191 kfilter_byname_sys(const char *name) 192 { 193 int i; 194 195 KASSERT(rw_lock_held(&kqueue_filter_lock)); 196 197 for (i = 0; sys_kfilters[i].name != NULL; i++) { 198 if (strcmp(name, sys_kfilters[i].name) == 0) 199 return &sys_kfilters[i]; 200 } 201 return NULL; 202 } 203 204 static struct kfilter * 205 kfilter_byname_user(const char *name) 206 { 207 int i; 208 209 KASSERT(rw_lock_held(&kqueue_filter_lock)); 210 211 /* user filter slots have a NULL name if previously deregistered */ 212 for (i = 0; i < user_kfilterc ; i++) { 213 if (user_kfilters[i].name != NULL && 214 strcmp(name, user_kfilters[i].name) == 0) 215 return &user_kfilters[i]; 216 } 217 return NULL; 218 } 219 220 static struct kfilter * 221 kfilter_byname(const char *name) 222 { 223 struct kfilter *kfilter; 224 225 KASSERT(rw_lock_held(&kqueue_filter_lock)); 226 227 if ((kfilter = kfilter_byname_sys(name)) != NULL) 228 return kfilter; 229 230 return kfilter_byname_user(name); 231 } 232 233 /* 234 * Find kfilter entry by filter id, or NULL if not found. 235 * Assumes entries are indexed in filter id order, for speed. 236 */ 237 static struct kfilter * 238 kfilter_byfilter(uint32_t filter) 239 { 240 struct kfilter *kfilter; 241 242 KASSERT(rw_lock_held(&kqueue_filter_lock)); 243 244 if (filter < EVFILT_SYSCOUNT) /* it's a system filter */ 245 kfilter = &sys_kfilters[filter]; 246 else if (user_kfilters != NULL && 247 filter < EVFILT_SYSCOUNT + user_kfilterc) 248 /* it's a user filter */ 249 kfilter = &user_kfilters[filter - EVFILT_SYSCOUNT]; 250 else 251 return (NULL); /* out of range */ 252 KASSERT(kfilter->filter == filter); /* sanity check! */ 253 return (kfilter); 254 } 255 256 /* 257 * Register a new kfilter. Stores the entry in user_kfilters. 258 * Returns 0 if operation succeeded, or an appropriate errno(2) otherwise. 259 * If retfilter != NULL, the new filterid is returned in it. 260 */ 261 int 262 kfilter_register(const char *name, const struct filterops *filtops, 263 int *retfilter) 264 { 265 struct kfilter *kfilter; 266 size_t len; 267 int i; 268 269 if (name == NULL || name[0] == '\0' || filtops == NULL) 270 return (EINVAL); /* invalid args */ 271 272 rw_enter(&kqueue_filter_lock, RW_WRITER); 273 if (kfilter_byname(name) != NULL) { 274 rw_exit(&kqueue_filter_lock); 275 return (EEXIST); /* already exists */ 276 } 277 if (user_kfilterc > 0xffffffff - EVFILT_SYSCOUNT) { 278 rw_exit(&kqueue_filter_lock); 279 return (EINVAL); /* too many */ 280 } 281 282 for (i = 0; i < user_kfilterc; i++) { 283 kfilter = &user_kfilters[i]; 284 if (kfilter->name == NULL) { 285 /* Previously deregistered slot. Reuse. */ 286 goto reuse; 287 } 288 } 289 290 /* check if need to grow user_kfilters */ 291 if (user_kfilterc + 1 > user_kfiltermaxc) { 292 /* Grow in KFILTER_EXTENT chunks. */ 293 user_kfiltermaxc += KFILTER_EXTENT; 294 len = user_kfiltermaxc * sizeof(struct filter *); 295 kfilter = kmem_alloc(len, KM_SLEEP); 296 memset((char *)kfilter + user_kfiltersz, 0, len - user_kfiltersz); 297 if (user_kfilters != NULL) { 298 memcpy(kfilter, user_kfilters, user_kfiltersz); 299 kmem_free(user_kfilters, user_kfiltersz); 300 } 301 user_kfiltersz = len; 302 user_kfilters = kfilter; 303 } 304 /* Adding new slot */ 305 kfilter = &user_kfilters[user_kfilterc++]; 306 reuse: 307 kfilter->namelen = strlen(name) + 1; 308 kfilter->name = kmem_alloc(kfilter->namelen, KM_SLEEP); 309 memcpy(__UNCONST(kfilter->name), name, kfilter->namelen); 310 311 kfilter->filter = (kfilter - user_kfilters) + EVFILT_SYSCOUNT; 312 313 kfilter->filtops = kmem_alloc(sizeof(*filtops), KM_SLEEP); 314 memcpy(__UNCONST(kfilter->filtops), filtops, sizeof(*filtops)); 315 316 if (retfilter != NULL) 317 *retfilter = kfilter->filter; 318 rw_exit(&kqueue_filter_lock); 319 320 return (0); 321 } 322 323 /* 324 * Unregister a kfilter previously registered with kfilter_register. 325 * This retains the filter id, but clears the name and frees filtops (filter 326 * operations), so that the number isn't reused during a boot. 327 * Returns 0 if operation succeeded, or an appropriate errno(2) otherwise. 328 */ 329 int 330 kfilter_unregister(const char *name) 331 { 332 struct kfilter *kfilter; 333 334 if (name == NULL || name[0] == '\0') 335 return (EINVAL); /* invalid name */ 336 337 rw_enter(&kqueue_filter_lock, RW_WRITER); 338 if (kfilter_byname_sys(name) != NULL) { 339 rw_exit(&kqueue_filter_lock); 340 return (EINVAL); /* can't detach system filters */ 341 } 342 343 kfilter = kfilter_byname_user(name); 344 if (kfilter == NULL) { 345 rw_exit(&kqueue_filter_lock); 346 return (ENOENT); 347 } 348 if (kfilter->refcnt != 0) { 349 rw_exit(&kqueue_filter_lock); 350 return (EBUSY); 351 } 352 353 /* Cast away const (but we know it's safe. */ 354 kmem_free(__UNCONST(kfilter->name), kfilter->namelen); 355 kfilter->name = NULL; /* mark as `not implemented' */ 356 357 if (kfilter->filtops != NULL) { 358 /* Cast away const (but we know it's safe. */ 359 kmem_free(__UNCONST(kfilter->filtops), 360 sizeof(*kfilter->filtops)); 361 kfilter->filtops = NULL; /* mark as `not implemented' */ 362 } 363 rw_exit(&kqueue_filter_lock); 364 365 return (0); 366 } 367 368 369 /* 370 * Filter attach method for EVFILT_READ and EVFILT_WRITE on normal file 371 * descriptors. Calls fileops kqfilter method for given file descriptor. 372 */ 373 static int 374 filt_fileattach(struct knote *kn) 375 { 376 file_t *fp; 377 378 fp = kn->kn_obj; 379 380 return (*fp->f_ops->fo_kqfilter)(fp, kn); 381 } 382 383 /* 384 * Filter detach method for EVFILT_READ on kqueue descriptor. 385 */ 386 static void 387 filt_kqdetach(struct knote *kn) 388 { 389 struct kqueue *kq; 390 391 kq = ((file_t *)kn->kn_obj)->f_data; 392 393 mutex_spin_enter(&kq->kq_lock); 394 SLIST_REMOVE(&kq->kq_sel.sel_klist, kn, knote, kn_selnext); 395 mutex_spin_exit(&kq->kq_lock); 396 } 397 398 /* 399 * Filter event method for EVFILT_READ on kqueue descriptor. 400 */ 401 /*ARGSUSED*/ 402 static int 403 filt_kqueue(struct knote *kn, long hint) 404 { 405 struct kqueue *kq; 406 int rv; 407 408 kq = ((file_t *)kn->kn_obj)->f_data; 409 410 if (hint != NOTE_SUBMIT) 411 mutex_spin_enter(&kq->kq_lock); 412 kn->kn_data = kq->kq_count; 413 rv = (kn->kn_data > 0); 414 if (hint != NOTE_SUBMIT) 415 mutex_spin_exit(&kq->kq_lock); 416 417 return rv; 418 } 419 420 /* 421 * Filter attach method for EVFILT_PROC. 422 */ 423 static int 424 filt_procattach(struct knote *kn) 425 { 426 struct proc *p, *curp; 427 struct lwp *curl; 428 429 curl = curlwp; 430 curp = curl->l_proc; 431 432 mutex_enter(&proclist_lock); 433 p = p_find(kn->kn_id, PFIND_LOCKED); 434 if (p == NULL) { 435 mutex_exit(&proclist_lock); 436 return ESRCH; 437 } 438 439 /* 440 * Fail if it's not owned by you, or the last exec gave us 441 * setuid/setgid privs (unless you're root). 442 */ 443 mutex_enter(&p->p_mutex); 444 mutex_exit(&proclist_lock); 445 if (kauth_authorize_process(curl->l_cred, KAUTH_PROCESS_KEVENT_FILTER, 446 p, NULL, NULL, NULL) != 0) { 447 mutex_exit(&p->p_mutex); 448 return EACCES; 449 } 450 451 kn->kn_obj = p; 452 kn->kn_flags |= EV_CLEAR; /* automatically set */ 453 454 /* 455 * internal flag indicating registration done by kernel 456 */ 457 if (kn->kn_flags & EV_FLAG1) { 458 kn->kn_data = kn->kn_sdata; /* ppid */ 459 kn->kn_fflags = NOTE_CHILD; 460 kn->kn_flags &= ~EV_FLAG1; 461 } 462 SLIST_INSERT_HEAD(&p->p_klist, kn, kn_selnext); 463 mutex_exit(&p->p_mutex); 464 465 return 0; 466 } 467 468 /* 469 * Filter detach method for EVFILT_PROC. 470 * 471 * The knote may be attached to a different process, which may exit, 472 * leaving nothing for the knote to be attached to. So when the process 473 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so 474 * it will be deleted when read out. However, as part of the knote deletion, 475 * this routine is called, so a check is needed to avoid actually performing 476 * a detach, because the original process might not exist any more. 477 */ 478 static void 479 filt_procdetach(struct knote *kn) 480 { 481 struct proc *p; 482 483 if (kn->kn_status & KN_DETACHED) 484 return; 485 486 p = kn->kn_obj; 487 488 mutex_enter(&p->p_mutex); 489 SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext); 490 mutex_exit(&p->p_mutex); 491 } 492 493 /* 494 * Filter event method for EVFILT_PROC. 495 */ 496 static int 497 filt_proc(struct knote *kn, long hint) 498 { 499 u_int event, fflag; 500 struct kevent kev; 501 struct kqueue *kq; 502 int error; 503 504 event = (u_int)hint & NOTE_PCTRLMASK; 505 kq = kn->kn_kq; 506 fflag = 0; 507 508 /* If the user is interested in this event, record it. */ 509 if (kn->kn_sfflags & event) 510 fflag |= event; 511 512 if (event == NOTE_EXIT) { 513 /* 514 * Process is gone, so flag the event as finished. 515 * 516 * Detach the knote from watched process and mark 517 * it as such. We can't leave this to kqueue_scan(), 518 * since the process might not exist by then. And we 519 * have to do this now, since psignal KNOTE() is called 520 * also for zombies and we might end up reading freed 521 * memory if the kevent would already be picked up 522 * and knote g/c'ed. 523 */ 524 filt_procdetach(kn); 525 526 mutex_spin_enter(&kq->kq_lock); 527 kn->kn_status |= KN_DETACHED; 528 /* Mark as ONESHOT, so that the knote it g/c'ed when read */ 529 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 530 kn->kn_fflags |= fflag; 531 mutex_spin_exit(&kq->kq_lock); 532 533 return 1; 534 } 535 536 mutex_spin_enter(&kq->kq_lock); 537 if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) { 538 /* 539 * Process forked, and user wants to track the new process, 540 * so attach a new knote to it, and immediately report an 541 * event with the parent's pid. Register knote with new 542 * process. 543 */ 544 kev.ident = hint & NOTE_PDATAMASK; /* pid */ 545 kev.filter = kn->kn_filter; 546 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1; 547 kev.fflags = kn->kn_sfflags; 548 kev.data = kn->kn_id; /* parent */ 549 kev.udata = kn->kn_kevent.udata; /* preserve udata */ 550 mutex_spin_exit(&kq->kq_lock); 551 error = kqueue_register(kq, &kev); 552 mutex_spin_enter(&kq->kq_lock); 553 if (error != 0) 554 kn->kn_fflags |= NOTE_TRACKERR; 555 } 556 kn->kn_fflags |= fflag; 557 fflag = kn->kn_fflags; 558 mutex_spin_exit(&kq->kq_lock); 559 560 return fflag != 0; 561 } 562 563 static void 564 filt_timerexpire(void *knx) 565 { 566 struct knote *kn = knx; 567 int tticks; 568 569 mutex_enter(&kqueue_misc_lock); 570 kn->kn_data++; 571 knote_activate(kn); 572 if ((kn->kn_flags & EV_ONESHOT) == 0) { 573 tticks = mstohz(kn->kn_sdata); 574 callout_schedule((callout_t *)kn->kn_hook, tticks); 575 } 576 mutex_exit(&kqueue_misc_lock); 577 } 578 579 /* 580 * data contains amount of time to sleep, in milliseconds 581 */ 582 static int 583 filt_timerattach(struct knote *kn) 584 { 585 callout_t *calloutp; 586 struct kqueue *kq; 587 int tticks; 588 589 tticks = mstohz(kn->kn_sdata); 590 591 /* if the supplied value is under our resolution, use 1 tick */ 592 if (tticks == 0) { 593 if (kn->kn_sdata == 0) 594 return EINVAL; 595 tticks = 1; 596 } 597 598 if (atomic_inc_uint_nv(&kq_ncallouts) >= kq_calloutmax || 599 (calloutp = kmem_alloc(sizeof(*calloutp), KM_NOSLEEP)) == NULL) { 600 atomic_dec_uint(&kq_ncallouts); 601 return ENOMEM; 602 } 603 callout_init(calloutp, CALLOUT_MPSAFE); 604 605 kq = kn->kn_kq; 606 mutex_spin_enter(&kq->kq_lock); 607 kn->kn_flags |= EV_CLEAR; /* automatically set */ 608 kn->kn_hook = calloutp; 609 mutex_spin_exit(&kq->kq_lock); 610 611 callout_reset(calloutp, tticks, filt_timerexpire, kn); 612 613 return (0); 614 } 615 616 static void 617 filt_timerdetach(struct knote *kn) 618 { 619 callout_t *calloutp; 620 621 calloutp = (callout_t *)kn->kn_hook; 622 callout_halt(calloutp, NULL); 623 callout_destroy(calloutp); 624 kmem_free(calloutp, sizeof(*calloutp)); 625 atomic_dec_uint(&kq_ncallouts); 626 } 627 628 static int 629 filt_timer(struct knote *kn, long hint) 630 { 631 int rv; 632 633 mutex_enter(&kqueue_misc_lock); 634 rv = (kn->kn_data != 0); 635 mutex_exit(&kqueue_misc_lock); 636 637 return rv; 638 } 639 640 /* 641 * filt_seltrue: 642 * 643 * This filter "event" routine simulates seltrue(). 644 */ 645 int 646 filt_seltrue(struct knote *kn, long hint) 647 { 648 649 /* 650 * We don't know how much data can be read/written, 651 * but we know that it *can* be. This is about as 652 * good as select/poll does as well. 653 */ 654 kn->kn_data = 0; 655 return (1); 656 } 657 658 /* 659 * This provides full kqfilter entry for device switch tables, which 660 * has same effect as filter using filt_seltrue() as filter method. 661 */ 662 static void 663 filt_seltruedetach(struct knote *kn) 664 { 665 /* Nothing to do */ 666 } 667 668 const struct filterops seltrue_filtops = 669 { 1, NULL, filt_seltruedetach, filt_seltrue }; 670 671 int 672 seltrue_kqfilter(dev_t dev, struct knote *kn) 673 { 674 switch (kn->kn_filter) { 675 case EVFILT_READ: 676 case EVFILT_WRITE: 677 kn->kn_fop = &seltrue_filtops; 678 break; 679 default: 680 return (EINVAL); 681 } 682 683 /* Nothing more to do */ 684 return (0); 685 } 686 687 /* 688 * kqueue(2) system call. 689 */ 690 int 691 sys_kqueue(struct lwp *l, const void *v, register_t *retval) 692 { 693 struct kqueue *kq; 694 file_t *fp; 695 int fd, error; 696 697 if ((error = fd_allocfile(&fp, &fd)) != 0) 698 return error; 699 fp->f_flag = FREAD | FWRITE; 700 fp->f_type = DTYPE_KQUEUE; 701 fp->f_ops = &kqueueops; 702 kq = kmem_zalloc(sizeof(*kq), KM_SLEEP); 703 mutex_init(&kq->kq_lock, MUTEX_DEFAULT, IPL_SCHED); 704 cv_init(&kq->kq_cv, "kqueue"); 705 selinit(&kq->kq_sel); 706 TAILQ_INIT(&kq->kq_head); 707 fp->f_data = kq; 708 *retval = fd; 709 kq->kq_fdp = curlwp->l_fd; 710 fd_affix(curproc, fp, fd); 711 return error; 712 } 713 714 /* 715 * kevent(2) system call. 716 */ 717 static int 718 kevent_fetch_changes(void *private, const struct kevent *changelist, 719 struct kevent *changes, size_t index, int n) 720 { 721 722 return copyin(changelist + index, changes, n * sizeof(*changes)); 723 } 724 725 static int 726 kevent_put_events(void *private, struct kevent *events, 727 struct kevent *eventlist, size_t index, int n) 728 { 729 730 return copyout(events, eventlist + index, n * sizeof(*events)); 731 } 732 733 static const struct kevent_ops kevent_native_ops = { 734 keo_private: NULL, 735 keo_fetch_timeout: copyin, 736 keo_fetch_changes: kevent_fetch_changes, 737 keo_put_events: kevent_put_events, 738 }; 739 740 int 741 sys_kevent(struct lwp *l, const struct sys_kevent_args *uap, register_t *retval) 742 { 743 /* { 744 syscallarg(int) fd; 745 syscallarg(const struct kevent *) changelist; 746 syscallarg(size_t) nchanges; 747 syscallarg(struct kevent *) eventlist; 748 syscallarg(size_t) nevents; 749 syscallarg(const struct timespec *) timeout; 750 } */ 751 752 return kevent1(retval, SCARG(uap, fd), SCARG(uap, changelist), 753 SCARG(uap, nchanges), SCARG(uap, eventlist), SCARG(uap, nevents), 754 SCARG(uap, timeout), &kevent_native_ops); 755 } 756 757 int 758 kevent1(register_t *retval, int fd, 759 const struct kevent *changelist, size_t nchanges, 760 struct kevent *eventlist, size_t nevents, 761 const struct timespec *timeout, 762 const struct kevent_ops *keops) 763 { 764 struct kevent *kevp; 765 struct kqueue *kq; 766 struct timespec ts; 767 size_t i, n, ichange; 768 int nerrors, error; 769 struct kevent kevbuf[8]; /* approx 300 bytes on 64-bit */ 770 file_t *fp; 771 772 /* check that we're dealing with a kq */ 773 fp = fd_getfile(fd); 774 if (fp == NULL) 775 return (EBADF); 776 777 if (fp->f_type != DTYPE_KQUEUE) { 778 fd_putfile(fd); 779 return (EBADF); 780 } 781 782 if (timeout != NULL) { 783 error = (*keops->keo_fetch_timeout)(timeout, &ts, sizeof(ts)); 784 if (error) 785 goto done; 786 timeout = &ts; 787 } 788 789 kq = (struct kqueue *)fp->f_data; 790 nerrors = 0; 791 ichange = 0; 792 793 /* traverse list of events to register */ 794 while (nchanges > 0) { 795 n = MIN(nchanges, __arraycount(kevbuf)); 796 error = (*keops->keo_fetch_changes)(keops->keo_private, 797 changelist, kevbuf, ichange, n); 798 if (error) 799 goto done; 800 for (i = 0; i < n; i++) { 801 kevp = &kevbuf[i]; 802 kevp->flags &= ~EV_SYSFLAGS; 803 /* register each knote */ 804 error = kqueue_register(kq, kevp); 805 if (error) { 806 if (nevents != 0) { 807 kevp->flags = EV_ERROR; 808 kevp->data = error; 809 error = (*keops->keo_put_events) 810 (keops->keo_private, kevp, 811 eventlist, nerrors, 1); 812 if (error) 813 goto done; 814 nevents--; 815 nerrors++; 816 } else { 817 goto done; 818 } 819 } 820 } 821 nchanges -= n; /* update the results */ 822 ichange += n; 823 } 824 if (nerrors) { 825 *retval = nerrors; 826 error = 0; 827 goto done; 828 } 829 830 /* actually scan through the events */ 831 error = kqueue_scan(fp, nevents, eventlist, timeout, retval, keops, 832 kevbuf, __arraycount(kevbuf)); 833 done: 834 fd_putfile(fd); 835 return (error); 836 } 837 838 /* 839 * Register a given kevent kev onto the kqueue 840 */ 841 static int 842 kqueue_register(struct kqueue *kq, struct kevent *kev) 843 { 844 struct kfilter *kfilter; 845 filedesc_t *fdp; 846 file_t *fp; 847 fdfile_t *ff; 848 struct knote *kn, *newkn; 849 struct klist *list; 850 int error, fd, rv; 851 852 fdp = kq->kq_fdp; 853 fp = NULL; 854 kn = NULL; 855 error = 0; 856 fd = 0; 857 858 newkn = kmem_zalloc(sizeof(*newkn), KM_SLEEP); 859 860 rw_enter(&kqueue_filter_lock, RW_READER); 861 kfilter = kfilter_byfilter(kev->filter); 862 if (kfilter == NULL || kfilter->filtops == NULL) { 863 /* filter not found nor implemented */ 864 rw_exit(&kqueue_filter_lock); 865 kmem_free(newkn, sizeof(*newkn)); 866 return (EINVAL); 867 } 868 869 mutex_enter(&fdp->fd_lock); 870 871 /* search if knote already exists */ 872 if (kfilter->filtops->f_isfd) { 873 /* monitoring a file descriptor */ 874 fd = kev->ident; 875 if ((fp = fd_getfile(fd)) == NULL) { 876 mutex_exit(&fdp->fd_lock); 877 rw_exit(&kqueue_filter_lock); 878 kmem_free(newkn, sizeof(*newkn)); 879 return EBADF; 880 } 881 ff = fdp->fd_ofiles[fd]; 882 if (fd <= fdp->fd_lastkqfile) { 883 SLIST_FOREACH(kn, &ff->ff_knlist, kn_link) { 884 if (kq == kn->kn_kq && 885 kev->filter == kn->kn_filter) 886 break; 887 } 888 } 889 } else { 890 /* 891 * not monitoring a file descriptor, so 892 * lookup knotes in internal hash table 893 */ 894 if (fdp->fd_knhashmask != 0) { 895 list = &fdp->fd_knhash[ 896 KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)]; 897 SLIST_FOREACH(kn, list, kn_link) { 898 if (kev->ident == kn->kn_id && 899 kq == kn->kn_kq && 900 kev->filter == kn->kn_filter) 901 break; 902 } 903 } 904 } 905 906 /* 907 * kn now contains the matching knote, or NULL if no match 908 */ 909 if (kev->flags & EV_ADD) { 910 if (kn == NULL) { 911 /* create new knote */ 912 kn = newkn; 913 newkn = NULL; 914 kn->kn_obj = fp; 915 kn->kn_kq = kq; 916 kn->kn_fop = kfilter->filtops; 917 kn->kn_kfilter = kfilter; 918 kn->kn_sfflags = kev->fflags; 919 kn->kn_sdata = kev->data; 920 kev->fflags = 0; 921 kev->data = 0; 922 kn->kn_kevent = *kev; 923 924 /* 925 * apply reference count to knote structure, and 926 * do not release it at the end of this routine. 927 */ 928 fp = NULL; 929 930 if (!kn->kn_fop->f_isfd) { 931 /* 932 * If knote is not on an fd, store on 933 * internal hash table. 934 */ 935 if (fdp->fd_knhashmask == 0) { 936 /* XXXAD can block with fd_lock held */ 937 fdp->fd_knhash = hashinit(KN_HASHSIZE, 938 HASH_LIST, M_KEVENT, M_WAITOK, 939 &fdp->fd_knhashmask); 940 } 941 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, 942 fdp->fd_knhashmask)]; 943 } else { 944 /* Otherwise, knote is on an fd. */ 945 list = (struct klist *) 946 &fdp->fd_ofiles[kn->kn_id]->ff_knlist; 947 if ((int)kn->kn_id > fdp->fd_lastkqfile) 948 fdp->fd_lastkqfile = kn->kn_id; 949 } 950 SLIST_INSERT_HEAD(list, kn, kn_link); 951 952 KERNEL_LOCK(1, NULL); /* XXXSMP */ 953 error = (*kfilter->filtops->f_attach)(kn); 954 KERNEL_UNLOCK_ONE(NULL); /* XXXSMP */ 955 if (error != 0) { 956 /* knote_detach() drops fdp->fd_lock */ 957 knote_detach(kn, fdp, false); 958 goto done; 959 } 960 atomic_inc_uint(&kfilter->refcnt); 961 } else { 962 /* 963 * The user may change some filter values after the 964 * initial EV_ADD, but doing so will not reset any 965 * filter which have already been triggered. 966 */ 967 kn->kn_sfflags = kev->fflags; 968 kn->kn_sdata = kev->data; 969 kn->kn_kevent.udata = kev->udata; 970 } 971 KERNEL_LOCK(1, NULL); /* XXXSMP */ 972 rv = (*kn->kn_fop->f_event)(kn, 0); 973 KERNEL_UNLOCK_ONE(NULL); /* XXXSMP */ 974 if (rv) 975 knote_activate(kn); 976 } else { 977 if (kn == NULL) { 978 error = ENOENT; 979 mutex_exit(&fdp->fd_lock); 980 goto done; 981 } 982 if (kev->flags & EV_DELETE) { 983 /* knote_detach() drops fdp->fd_lock */ 984 knote_detach(kn, fdp, true); 985 goto done; 986 } 987 } 988 989 /* disable knote */ 990 if ((kev->flags & EV_DISABLE)) { 991 mutex_spin_enter(&kq->kq_lock); 992 if ((kn->kn_status & KN_DISABLED) == 0) 993 kn->kn_status |= KN_DISABLED; 994 mutex_spin_exit(&kq->kq_lock); 995 } 996 997 /* enable knote */ 998 if ((kev->flags & EV_ENABLE)) { 999 knote_enqueue(kn); 1000 } 1001 mutex_exit(&fdp->fd_lock); 1002 done: 1003 rw_exit(&kqueue_filter_lock); 1004 if (newkn != NULL) 1005 kmem_free(newkn, sizeof(*newkn)); 1006 if (fp != NULL) 1007 fd_putfile(fd); 1008 return (error); 1009 } 1010 1011 #if defined(DEBUG) 1012 static void 1013 kq_check(struct kqueue *kq) 1014 { 1015 const struct knote *kn; 1016 int count; 1017 int nmarker; 1018 1019 KASSERT(mutex_owned(&kq->kq_lock)); 1020 KASSERT(kq->kq_count >= 0); 1021 1022 count = 0; 1023 nmarker = 0; 1024 TAILQ_FOREACH(kn, &kq->kq_head, kn_tqe) { 1025 if ((kn->kn_status & (KN_MARKER | KN_QUEUED)) == 0) { 1026 panic("%s: kq=%p kn=%p inconsist 1", __func__, kq, kn); 1027 } 1028 if ((kn->kn_status & KN_MARKER) == 0) { 1029 if (kn->kn_kq != kq) { 1030 panic("%s: kq=%p kn=%p inconsist 2", 1031 __func__, kq, kn); 1032 } 1033 if ((kn->kn_status & KN_ACTIVE) == 0) { 1034 panic("%s: kq=%p kn=%p: not active", 1035 __func__, kq, kn); 1036 } 1037 count++; 1038 if (count > kq->kq_count) { 1039 goto bad; 1040 } 1041 } else { 1042 nmarker++; 1043 #if 0 1044 if (nmarker > 10000) { 1045 panic("%s: kq=%p too many markers: %d != %d, " 1046 "nmarker=%d", 1047 __func__, kq, kq->kq_count, count, nmarker); 1048 } 1049 #endif 1050 } 1051 } 1052 if (kq->kq_count != count) { 1053 bad: 1054 panic("%s: kq=%p inconsist 3: %d != %d, nmarker=%d", 1055 __func__, kq, kq->kq_count, count, nmarker); 1056 } 1057 } 1058 #else /* defined(DEBUG) */ 1059 #define kq_check(a) /* nothing */ 1060 #endif /* defined(DEBUG) */ 1061 1062 /* 1063 * Scan through the list of events on fp (for a maximum of maxevents), 1064 * returning the results in to ulistp. Timeout is determined by tsp; if 1065 * NULL, wait indefinitely, if 0 valued, perform a poll, otherwise wait 1066 * as appropriate. 1067 */ 1068 static int 1069 kqueue_scan(file_t *fp, size_t maxevents, struct kevent *ulistp, 1070 const struct timespec *tsp, register_t *retval, 1071 const struct kevent_ops *keops, struct kevent *kevbuf, 1072 size_t kevcnt) 1073 { 1074 struct kqueue *kq; 1075 struct kevent *kevp; 1076 struct timeval atv, sleeptv; 1077 struct knote *kn, *marker; 1078 size_t count, nkev, nevents; 1079 int timeout, error, rv; 1080 filedesc_t *fdp; 1081 1082 fdp = curlwp->l_fd; 1083 kq = fp->f_data; 1084 count = maxevents; 1085 nkev = nevents = error = 0; 1086 if (count == 0) { 1087 *retval = 0; 1088 return 0; 1089 } 1090 1091 if (tsp) { /* timeout supplied */ 1092 TIMESPEC_TO_TIMEVAL(&atv, tsp); 1093 if (inittimeleft(&atv, &sleeptv) == -1) { 1094 *retval = maxevents; 1095 return EINVAL; 1096 } 1097 timeout = tvtohz(&atv); 1098 if (timeout <= 0) 1099 timeout = -1; /* do poll */ 1100 } else { 1101 /* no timeout, wait forever */ 1102 timeout = 0; 1103 } 1104 1105 marker = kmem_zalloc(sizeof(*marker), KM_SLEEP); 1106 marker->kn_status = KN_MARKER; 1107 mutex_spin_enter(&kq->kq_lock); 1108 retry: 1109 kevp = kevbuf; 1110 if (kq->kq_count == 0) { 1111 if (timeout >= 0) { 1112 error = cv_timedwait_sig(&kq->kq_cv, 1113 &kq->kq_lock, timeout); 1114 if (error == 0) { 1115 if (tsp == NULL || (timeout = 1116 gettimeleft(&atv, &sleeptv)) > 0) 1117 goto retry; 1118 } else { 1119 /* don't restart after signals... */ 1120 if (error == ERESTART) 1121 error = EINTR; 1122 if (error == EWOULDBLOCK) 1123 error = 0; 1124 } 1125 } 1126 } else { 1127 /* mark end of knote list */ 1128 TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe); 1129 1130 while (count != 0) { 1131 kn = TAILQ_FIRST(&kq->kq_head); /* get next knote */ 1132 while ((kn->kn_status & KN_MARKER) != 0) { 1133 if (kn == marker) { 1134 /* it's our marker, stop */ 1135 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 1136 if (count < maxevents || (tsp != NULL && 1137 (timeout = gettimeleft(&atv, 1138 &sleeptv)) <= 0)) 1139 goto done; 1140 goto retry; 1141 } 1142 /* someone else's marker. */ 1143 kn = TAILQ_NEXT(kn, kn_tqe); 1144 } 1145 kq_check(kq); 1146 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 1147 kq->kq_count--; 1148 kn->kn_status &= ~KN_QUEUED; 1149 kq_check(kq); 1150 if (kn->kn_status & KN_DISABLED) { 1151 /* don't want disabled events */ 1152 continue; 1153 } 1154 if ((kn->kn_flags & EV_ONESHOT) == 0) { 1155 mutex_spin_exit(&kq->kq_lock); 1156 KERNEL_LOCK(1, NULL); /* XXXSMP */ 1157 rv = (*kn->kn_fop->f_event)(kn, 0); 1158 KERNEL_UNLOCK_ONE(NULL); /* XXXSMP */ 1159 mutex_spin_enter(&kq->kq_lock); 1160 /* Re-poll if note was re-enqueued. */ 1161 if ((kn->kn_status & KN_QUEUED) != 0) 1162 continue; 1163 if (rv == 0) { 1164 /* 1165 * non-ONESHOT event that hasn't 1166 * triggered again, so de-queue. 1167 */ 1168 kn->kn_status &= ~KN_ACTIVE; 1169 continue; 1170 } 1171 } 1172 /* XXXAD should be got from f_event if !oneshot. */ 1173 *kevp++ = kn->kn_kevent; 1174 nkev++; 1175 if (kn->kn_flags & EV_ONESHOT) { 1176 /* delete ONESHOT events after retrieval */ 1177 mutex_spin_exit(&kq->kq_lock); 1178 mutex_enter(&fdp->fd_lock); 1179 knote_detach(kn, fdp, true); 1180 mutex_spin_enter(&kq->kq_lock); 1181 } else if (kn->kn_flags & EV_CLEAR) { 1182 /* clear state after retrieval */ 1183 kn->kn_data = 0; 1184 kn->kn_fflags = 0; 1185 kn->kn_status &= ~KN_ACTIVE; 1186 } else { 1187 /* add event back on list */ 1188 kq_check(kq); 1189 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 1190 kq->kq_count++; 1191 kn->kn_status |= KN_QUEUED; 1192 kq_check(kq); 1193 } 1194 if (nkev == kevcnt) { 1195 /* do copyouts in kevcnt chunks */ 1196 mutex_spin_exit(&kq->kq_lock); 1197 error = (*keops->keo_put_events) 1198 (keops->keo_private, 1199 kevbuf, ulistp, nevents, nkev); 1200 mutex_spin_enter(&kq->kq_lock); 1201 nevents += nkev; 1202 nkev = 0; 1203 kevp = kevbuf; 1204 } 1205 count--; 1206 if (error != 0 || count == 0) { 1207 /* remove marker */ 1208 TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe); 1209 break; 1210 } 1211 } 1212 } 1213 done: 1214 mutex_spin_exit(&kq->kq_lock); 1215 if (marker != NULL) 1216 kmem_free(marker, sizeof(*marker)); 1217 if (nkev != 0) { 1218 /* copyout remaining events */ 1219 error = (*keops->keo_put_events)(keops->keo_private, 1220 kevbuf, ulistp, nevents, nkev); 1221 } 1222 *retval = maxevents - count; 1223 1224 return error; 1225 } 1226 1227 /* 1228 * fileops ioctl method for a kqueue descriptor. 1229 * 1230 * Two ioctls are currently supported. They both use struct kfilter_mapping: 1231 * KFILTER_BYNAME find name for filter, and return result in 1232 * name, which is of size len. 1233 * KFILTER_BYFILTER find filter for name. len is ignored. 1234 */ 1235 /*ARGSUSED*/ 1236 static int 1237 kqueue_ioctl(file_t *fp, u_long com, void *data) 1238 { 1239 struct kfilter_mapping *km; 1240 const struct kfilter *kfilter; 1241 char *name; 1242 int error; 1243 1244 km = data; 1245 error = 0; 1246 name = kmem_alloc(KFILTER_MAXNAME, KM_SLEEP); 1247 1248 switch (com) { 1249 case KFILTER_BYFILTER: /* convert filter -> name */ 1250 rw_enter(&kqueue_filter_lock, RW_READER); 1251 kfilter = kfilter_byfilter(km->filter); 1252 if (kfilter != NULL) { 1253 strlcpy(name, kfilter->name, KFILTER_MAXNAME); 1254 rw_exit(&kqueue_filter_lock); 1255 error = copyoutstr(name, km->name, km->len, NULL); 1256 } else { 1257 rw_exit(&kqueue_filter_lock); 1258 error = ENOENT; 1259 } 1260 break; 1261 1262 case KFILTER_BYNAME: /* convert name -> filter */ 1263 error = copyinstr(km->name, name, KFILTER_MAXNAME, NULL); 1264 if (error) { 1265 break; 1266 } 1267 rw_enter(&kqueue_filter_lock, RW_READER); 1268 kfilter = kfilter_byname(name); 1269 if (kfilter != NULL) 1270 km->filter = kfilter->filter; 1271 else 1272 error = ENOENT; 1273 rw_exit(&kqueue_filter_lock); 1274 break; 1275 1276 default: 1277 error = ENOTTY; 1278 break; 1279 1280 } 1281 kmem_free(name, KFILTER_MAXNAME); 1282 return (error); 1283 } 1284 1285 /* 1286 * fileops fcntl method for a kqueue descriptor. 1287 */ 1288 static int 1289 kqueue_fcntl(file_t *fp, u_int com, void *data) 1290 { 1291 1292 return (ENOTTY); 1293 } 1294 1295 /* 1296 * fileops poll method for a kqueue descriptor. 1297 * Determine if kqueue has events pending. 1298 */ 1299 static int 1300 kqueue_poll(file_t *fp, int events) 1301 { 1302 struct kqueue *kq; 1303 int revents; 1304 1305 kq = fp->f_data; 1306 1307 revents = 0; 1308 if (events & (POLLIN | POLLRDNORM)) { 1309 mutex_spin_enter(&kq->kq_lock); 1310 if (kq->kq_count != 0) { 1311 revents |= events & (POLLIN | POLLRDNORM); 1312 } else { 1313 selrecord(curlwp, &kq->kq_sel); 1314 } 1315 kq_check(kq); 1316 mutex_spin_exit(&kq->kq_lock); 1317 } 1318 1319 return revents; 1320 } 1321 1322 /* 1323 * fileops stat method for a kqueue descriptor. 1324 * Returns dummy info, with st_size being number of events pending. 1325 */ 1326 static int 1327 kqueue_stat(file_t *fp, struct stat *st) 1328 { 1329 struct kqueue *kq; 1330 1331 kq = fp->f_data; 1332 1333 memset(st, 0, sizeof(*st)); 1334 st->st_size = kq->kq_count; 1335 st->st_blksize = sizeof(struct kevent); 1336 st->st_mode = S_IFIFO; 1337 1338 return 0; 1339 } 1340 1341 static void 1342 kqueue_doclose(struct kqueue *kq, struct klist *list, int fd) 1343 { 1344 struct knote *kn; 1345 filedesc_t *fdp; 1346 1347 fdp = kq->kq_fdp; 1348 1349 KASSERT(mutex_owned(&fdp->fd_lock)); 1350 1351 for (kn = SLIST_FIRST(list); kn != NULL;) { 1352 if (kq != kn->kn_kq) { 1353 kn = SLIST_NEXT(kn, kn_link); 1354 continue; 1355 } 1356 knote_detach(kn, fdp, true); 1357 mutex_enter(&fdp->fd_lock); 1358 kn = SLIST_FIRST(list); 1359 } 1360 } 1361 1362 1363 /* 1364 * fileops close method for a kqueue descriptor. 1365 */ 1366 static int 1367 kqueue_close(file_t *fp) 1368 { 1369 struct kqueue *kq; 1370 filedesc_t *fdp; 1371 fdfile_t *ff; 1372 int i; 1373 1374 kq = fp->f_data; 1375 fdp = curlwp->l_fd; 1376 1377 mutex_enter(&fdp->fd_lock); 1378 for (i = 0; i <= fdp->fd_lastkqfile; i++) { 1379 if ((ff = fdp->fd_ofiles[i]) == NULL) 1380 continue; 1381 kqueue_doclose(kq, (struct klist *)&ff->ff_knlist, i); 1382 } 1383 if (fdp->fd_knhashmask != 0) { 1384 for (i = 0; i < fdp->fd_knhashmask + 1; i++) { 1385 kqueue_doclose(kq, &fdp->fd_knhash[i], -1); 1386 } 1387 } 1388 mutex_exit(&fdp->fd_lock); 1389 1390 KASSERT(kq->kq_count == 0); 1391 mutex_destroy(&kq->kq_lock); 1392 cv_destroy(&kq->kq_cv); 1393 seldestroy(&kq->kq_sel); 1394 kmem_free(kq, sizeof(*kq)); 1395 fp->f_data = NULL; 1396 1397 return (0); 1398 } 1399 1400 /* 1401 * struct fileops kqfilter method for a kqueue descriptor. 1402 * Event triggered when monitored kqueue changes. 1403 */ 1404 static int 1405 kqueue_kqfilter(file_t *fp, struct knote *kn) 1406 { 1407 struct kqueue *kq; 1408 filedesc_t *fdp; 1409 1410 kq = ((file_t *)kn->kn_obj)->f_data; 1411 1412 KASSERT(fp == kn->kn_obj); 1413 1414 if (kn->kn_filter != EVFILT_READ) 1415 return 1; 1416 1417 kn->kn_fop = &kqread_filtops; 1418 fdp = curlwp->l_fd; 1419 mutex_enter(&kq->kq_lock); 1420 SLIST_INSERT_HEAD(&kq->kq_sel.sel_klist, kn, kn_selnext); 1421 mutex_exit(&kq->kq_lock); 1422 1423 return 0; 1424 } 1425 1426 1427 /* 1428 * Walk down a list of knotes, activating them if their event has 1429 * triggered. The caller's object lock (e.g. device driver lock) 1430 * must be held. 1431 */ 1432 void 1433 knote(struct klist *list, long hint) 1434 { 1435 struct knote *kn; 1436 1437 SLIST_FOREACH(kn, list, kn_selnext) { 1438 if ((*kn->kn_fop->f_event)(kn, hint)) 1439 knote_activate(kn); 1440 } 1441 } 1442 1443 /* 1444 * Remove all knotes referencing a specified fd 1445 */ 1446 void 1447 knote_fdclose(int fd) 1448 { 1449 struct klist *list; 1450 struct knote *kn; 1451 filedesc_t *fdp; 1452 1453 fdp = curlwp->l_fd; 1454 list = (struct klist *)&fdp->fd_ofiles[fd]->ff_knlist; 1455 mutex_enter(&fdp->fd_lock); 1456 while ((kn = SLIST_FIRST(list)) != NULL) { 1457 knote_detach(kn, fdp, true); 1458 mutex_enter(&fdp->fd_lock); 1459 } 1460 mutex_exit(&fdp->fd_lock); 1461 } 1462 1463 /* 1464 * Drop knote. Called with fdp->fd_lock held, and will drop before 1465 * returning. 1466 */ 1467 static void 1468 knote_detach(struct knote *kn, filedesc_t *fdp, bool dofop) 1469 { 1470 struct klist *list; 1471 struct kqueue *kq; 1472 1473 kq = kn->kn_kq; 1474 1475 KASSERT((kn->kn_status & KN_MARKER) == 0); 1476 KASSERT(mutex_owned(&fdp->fd_lock)); 1477 1478 /* Remove from monitored object. */ 1479 if (dofop) { 1480 KERNEL_LOCK(1, NULL); /* XXXSMP */ 1481 (*kn->kn_fop->f_detach)(kn); 1482 KERNEL_UNLOCK_ONE(NULL); /* XXXSMP */ 1483 } 1484 1485 /* Remove from descriptor table. */ 1486 if (kn->kn_fop->f_isfd) 1487 list = (struct klist *)&fdp->fd_ofiles[kn->kn_id]->ff_knlist; 1488 else 1489 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)]; 1490 1491 SLIST_REMOVE(list, kn, knote, kn_link); 1492 1493 /* Remove from kqueue. */ 1494 /* XXXAD should verify not in use by kqueue_scan. */ 1495 mutex_spin_enter(&kq->kq_lock); 1496 if ((kn->kn_status & KN_QUEUED) != 0) { 1497 kq_check(kq); 1498 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 1499 kn->kn_status &= ~KN_QUEUED; 1500 kq->kq_count--; 1501 kq_check(kq); 1502 } 1503 mutex_spin_exit(&kq->kq_lock); 1504 1505 mutex_exit(&fdp->fd_lock); 1506 if (kn->kn_fop->f_isfd) 1507 fd_putfile(kn->kn_id); 1508 atomic_dec_uint(&kn->kn_kfilter->refcnt); 1509 kmem_free(kn, sizeof(*kn)); 1510 } 1511 1512 /* 1513 * Queue new event for knote. 1514 */ 1515 static void 1516 knote_enqueue(struct knote *kn) 1517 { 1518 struct kqueue *kq; 1519 1520 KASSERT((kn->kn_status & KN_MARKER) == 0); 1521 1522 kq = kn->kn_kq; 1523 1524 mutex_spin_enter(&kq->kq_lock); 1525 if ((kn->kn_status & KN_DISABLED) != 0) { 1526 kn->kn_status &= ~KN_DISABLED; 1527 } 1528 if ((kn->kn_status & (KN_ACTIVE | KN_QUEUED)) == KN_ACTIVE) { 1529 kq_check(kq); 1530 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 1531 kn->kn_status |= KN_QUEUED; 1532 kq->kq_count++; 1533 kq_check(kq); 1534 cv_broadcast(&kq->kq_cv); 1535 selnotify(&kq->kq_sel, 0, NOTE_SUBMIT); 1536 } 1537 mutex_spin_exit(&kq->kq_lock); 1538 } 1539 /* 1540 * Queue new event for knote. 1541 */ 1542 static void 1543 knote_activate(struct knote *kn) 1544 { 1545 struct kqueue *kq; 1546 1547 KASSERT((kn->kn_status & KN_MARKER) == 0); 1548 1549 kq = kn->kn_kq; 1550 1551 mutex_spin_enter(&kq->kq_lock); 1552 kn->kn_status |= KN_ACTIVE; 1553 if ((kn->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) { 1554 kq_check(kq); 1555 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 1556 kn->kn_status |= KN_QUEUED; 1557 kq->kq_count++; 1558 kq_check(kq); 1559 cv_broadcast(&kq->kq_cv); 1560 selnotify(&kq->kq_sel, 0, NOTE_SUBMIT); 1561 } 1562 mutex_spin_exit(&kq->kq_lock); 1563 } 1564